Democracy by design or default? AI’s urban recommendations favor elites over public engagement

A groundbreaking study has cast a critical eye on how generative artificial intelligence is shaping perceptions of civic participation in urban spaces. The research, titled “Spaces for Democracy with Generative Artificial Intelligence: Public Architecture at Stake”, was published in the journal AI & Society. It explores the spatial narratives of democracy through outputs generated by ChatGPT-4o, analyzing the types of urban places it recommends for democratic engagement in Stockholm, Madrid, and Brussels.

Departing from the usual discourse around misinformation and bias, the study introduces a new conceptual lens: how generative AI influences public understanding of democratic participation through architecture and spatial suggestion. The findings are based on systematic prompt interactions with ChatGPT and reveal a narrowing of public imagination, often omitting spaces essential for protest, spontaneous assembly, or grassroots activism.

What kind of democratic spaces does AI recommend?

The study’s empirical core stems from a detailed content analysis of ChatGPT-4o responses to a series of structured prompts. When asked to list places in Stockholm that promote democracy or civic engagement, the AI consistently identified prominent institutional buildings and NGOs. These included the Swedish Parliament (Riksdag), Stockholm City Hall, ABF Stockholm (Workers' Educational Association), Folkets Hus (People’s House), and Kulturhuset Stadsteatern (The Culture House).

While these sites certainly facilitate representative and deliberative democracy, the AI showed a strong bias toward formalized, controlled environments. It emphasized institutions where access is mediated, physically, financially, or educationally, leaving out informal or spontaneous venues for public expression. Locations such as Sergels Torg or Mynttorget, which are historically vital for protests and large-scale public mobilization, were conspicuously absent.

This institutional emphasis extended to the kinds of participation recognized. Most of the places listed promoted educational lectures, organized debates, or youth programs, activities grounded in scheduled and regulated formats. The AI outputs largely overlooked the significance of routes, public squares, and outdoor green spaces that enable dynamic, movement-based demonstrations, offering a static view of civic life.

What are the spatial and political consequences of AI’s civic mapping?

Mapping data from the study revealed a strong clustering of AI-identified democratic spaces in central Stockholm - areas associated with high income and corporate presence. This centralization creates an unintentional yet troubling correlation between economic power and democratic legitimacy.

Drawing on the work of urban theorists like Manfredo Tafuri, David Harvey, and Chantal Mouffe, the research highlights the risk of reducing democratic engagement to formal institutions situated in economic hubs. This spatial bias reinforces the dominance of the urban elite in the democratic process and excludes suburban populations, particularly immigrants and working-class communities in areas like Rinkeby-Kista and Husby.

Moreover, grassroots organizations such as Megafonen, which have played pivotal roles in advocating for marginalized communities in Stockholm, were absent from ChatGPT’s responses. The AI also failed to mention labor union headquarters or political party offices, thereby erasing significant actors in Sweden’s democratic history.

From a democratic theory standpoint, this exclusion distorts the public sphere. As Mouffe argues, democracy thrives not on consensus but on visible, structured antagonism. Spaces like public squares are critical for manifesting dissent. Yet, AI-generated results portray a sanitized vision of democracy, minimizing conflict and elevating bureaucratic engagement.

How might this shape public perceptions of democracy?

The implications of the study extend far beyond Stockholm. Comparative prompts conducted on Brussels and Madrid revealed a similar tendency: ChatGPT listed headquarters of NGOs and think tanks rather than traditional public forums. In Madrid, entities like Club de Madrid and Fundación Alternativas were named, while in Brussels, democracy was personified through offices like International IDEA and the European Endowment for Democracy. In both cases, prominent public squares, political party hubs, and open protest grounds were omitted.

This shift suggests a deeper transformation in the way AI models internalize and project civic values. If the public increasingly depends on AI to interpret the civic landscape, then algorithmic blind spots could skew societal understanding of democratic engagement. The study warns that AI may inadvertently devalue the physical infrastructure of public space, an essential democratic asset, and subtly channel civic action into narrowly defined institutional tracks.

In doing so, AI risks amplifying the roles of elite actors and private organizations, while disempowering the everyday citizen’s role in shaping urban political life. More alarmingly, it could alter how cities are planned or funded, with investment flowing toward formal civic buildings while protest zones, parks, and community centers lose visibility in civic design dialogues.

Rethinking democracy with AI in the loop

The research concludes with a call for critical oversight. AI-generated frameworks for civic engagement must be cross-examined with local knowledge, participatory planning, and historical context. Urban space is more than infrastructure - it is a living, contested domain where political agency unfolds across class lines, spatial divides, and ideological battlegrounds.

Governments, planners, and technology developers are urged to reintroduce deliberative, spontaneous, and antagonistic forms of engagement into AI training datasets and outputs. As AI increasingly guides urban design, civic education, and decision-making, it is imperative that its frameworks do not narrow the democratic imagination.

In the age of generative AI, the future of democracy might depend not only on who participates, but on where they are allowed to stand, and which spaces AI recognizes as democratic ground. The architectural stakes, the study warns, are as high as the ideological ones.